| Commit message (Collapse) | Author | Age | Files | Lines |
| |
|
| |
|
|
|
|
| |
It was only used for scaling, it fails far too often and is only
faster on a limited selectoin of hardware.
|
| |
|
| |
|
| |
|
|
|
|
|
|
|
|
|
|
| |
Convert qsv_livav.[ch] to be built within libhb.
Convert other files to use qsv_libav.h in place of libavcodec/qsv.h
Don't attempt to build libav with QSV support (not available in unpatched release/11 branch).
QSV-accelerated decoding is broken, so disable it for the time being.
|
| |
|
|
|
|
| |
code commented out during a debug session got unintentionally committed
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
* sync: correct timestamp discontinuities in sync instead of reader
This patch passes discontinuity information through the pipeline till it
reaches sync.c. The timestamps are passed through the pipeline as read
and unmodified to sync.c (instead of attempting to correct
discontinuities in reader). In sync, when we see a discontinuity,
we know where the next timestamp should be based on the timestamp
and duration of the previous buffer (before the discontinuity). So
we calculate an "SCR" offset based on the timestamp after the
discontinuity and what we calculate it should be.
The old discontinuity handling code was broken due to the following.
The MPEG STD timing model relies heavily on the decoder having an STC
that is phase lock looped to the PCRs in the stream. When decoding a
broadcast stream, the decoder can count on the time measure between PCRs
using the STC to match to a high degree of accuracy.
I.e. STC - lastSTC == PCR - lastPCR. When a discontinuity occurs, the
decoder calculates a new PCR offset = PCR - STC. I.e. the offset is the
new PCR value minus what it would have been if there had been no
discontinuity.
The above does not work without a reliable STC, which we do not have.
We have been attempting to approximate one by avereraging the duration
of received packets and extrapolating an "STC" based on the last PTS and
the average packet duration. But this is highly variable and
unreliable.
* decavcodec: fix data type of next_pts
It needs to be double so that partial ticks are not lost
* deccc608sub: clarify comment
* sync: allow queueing more audio
Audio is small, and there is often a significant amount of audio in the
stream before the first video frame.
* sync: improve handling of damaged streams
When data is missing, the audio decoder was extrapolating timestamps
from the last pts before the error caused by the missing data which
caused sync issues.
Also, missing data can cause the video decoder to output a frame out of
order with the wrong scr sequence. Drop such frames.
|
| |
|
| |
|
| |
|
|
|
|
|
|
|
|
|
|
|
| |
sync.c was difficult to read, fragile, and prone to difficult to
diagnose bugs (see
https://forum.handbrake.fr/viewtopic.php?f=12&t=33147)
This rewrite simplifies the code, removes signals, locking and yield
that probably cause the above problem and is much more flexible. It
fixes a wider variety of timestamp issues than before and is much easier
to extend if other timestamp analysis is desired.
|
| |
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
| |
... instead of a 0 length buffer.
This fixes this issue:
https://forum.handbrake.fr/viewtopic.php?f=12&t=31959
Theora can create 0 length output. These 0 length frames indicate
duplicate frames. So we can't use 0 length buffers to indicate the end
of the stream.
git-svn-id: svn://svn.handbrake.fr/HandBrake/trunk@7143 b64f7644-9d1e-0410-96f1-a4d463321fa5
|
|
|
|
| |
git-svn-id: svn://svn.handbrake.fr/HandBrake/trunk@7119 b64f7644-9d1e-0410-96f1-a4d463321fa5
|
|
|
|
| |
git-svn-id: svn://svn.handbrake.fr/HandBrake/trunk@6852 b64f7644-9d1e-0410-96f1-a4d463321fa5
|
|
|
|
| |
git-svn-id: svn://svn.handbrake.fr/HandBrake/trunk@6613 b64f7644-9d1e-0410-96f1-a4d463321fa5
|
|
|
|
| |
git-svn-id: svn://svn.handbrake.fr/HandBrake/trunk@6485 b64f7644-9d1e-0410-96f1-a4d463321fa5
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
This new function has a couple advantages over the old one (which we
should phase out).
It does not require hb_job_t as a parameter, instead it uses hb_ui_geometry_t
which is a smaller and simpler struct. The entire job struct is overkill as
input to this function.
It returns an hb_image_t that fully describes the returned image instead of
just a uint8_t array. The caller does not have to make assumptions about image
size, line stide, or pixel format since hb_image_t specifies these things.
git-svn-id: svn://svn.handbrake.fr/HandBrake/trunk@6242 b64f7644-9d1e-0410-96f1-a4d463321fa5
|
|
|
|
| |
git-svn-id: svn://svn.handbrake.fr/HandBrake/trunk@6042 b64f7644-9d1e-0410-96f1-a4d463321fa5
|
|
|
|
|
|
|
|
|
| |
-1 is not a good value as a flag for invalid timestamps.
There are cases where small negative timestamps are useful.
So this eliminates a potential ambiguity.
git-svn-id: svn://svn.handbrake.fr/HandBrake/trunk@6001 b64f7644-9d1e-0410-96f1-a4d463321fa5
|
|
|
|
| |
git-svn-id: svn://svn.handbrake.fr/HandBrake/trunk@5954 b64f7644-9d1e-0410-96f1-a4d463321fa5
|
|
|
|
| |
git-svn-id: svn://svn.handbrake.fr/HandBrake/trunk@5905 b64f7644-9d1e-0410-96f1-a4d463321fa5
|
|
|
|
|
|
|
|
|
|
|
| |
An OpenCL SDK is no longer needed to build OpenCL support.
Note: as a result, the --enable-opencl configure option is removed.
Also, libOpenCL is no longer needed to run the application (it is still necessary to use OpenCL features, of course).
git-svn-id: svn://svn.handbrake.fr/HandBrake/trunk@5886 b64f7644-9d1e-0410-96f1-a4d463321fa5
|
|
|
|
|
|
|
|
|
|
|
| |
Filters were leaking buffers when a job is cancelled.
decavcodec could leak when job cancelled.
decavcodec leaked audio extradata in BSInfo
encavcodec and encavcodecaudio leaked AVCodecContext
sync leaked subtitle_sanitizer_t data
git-svn-id: svn://svn.handbrake.fr/HandBrake/trunk@5853 b64f7644-9d1e-0410-96f1-a4d463321fa5
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Patch originally by the Multicoreware Inc team, followed by improvements and fixes by Micheal Wootton from AMD Inc,
OpenCL:
This patch implements Bicubic Scaling in OpenCL.
Note that HandBrake currently uses Lanczos so the performance difference appears to be much more significant. We may offer an option of BiCubic in software later.
Bicubic scaling may appear a bit sharper than the equivalent Lanczos encode and may increase file size a bit. Quality may be better or worse depending on the scaling and content and personal preference towards sharpness.
When comparing performance with a custom HandBrake build that runs Software Bicubic to OpenCL Bicubic, performance increase is about 5~7% on average on a modern GPU.
Hardware Decode via DXVA:
We also have optional DXVA decoding which may come in useful for slower/lower end systems that have a capable GPU.
This is only available on input sources that use the libav decode path.
Most GPU hardware for decoding is designed for playback, so if you are running on a high end CPU, it will bottleneck the encode process.
Requires OpenCL 1.1 or later supporting GPU.
Front end changes and testing framework are not included in this patch. This will be resolved later.
Patch will be revised further before the UI is implemented.
git-svn-id: svn://svn.handbrake.fr/HandBrake/trunk@5792 b64f7644-9d1e-0410-96f1-a4d463321fa5
|
|
|
|
| |
git-svn-id: svn://svn.handbrake.fr/HandBrake/trunk@5737 b64f7644-9d1e-0410-96f1-a4d463321fa5
|
|
|
|
| |
git-svn-id: svn://svn.handbrake.fr/HandBrake/trunk@5620 b64f7644-9d1e-0410-96f1-a4d463321fa5
|
|
|
|
| |
git-svn-id: svn://svn.handbrake.fr/HandBrake/trunk@5318 b64f7644-9d1e-0410-96f1-a4d463321fa5
|
|
|
|
| |
git-svn-id: svn://svn.handbrake.fr/HandBrake/trunk@4963 b64f7644-9d1e-0410-96f1-a4d463321fa5
|
|
|
|
|
|
|
|
| |
memsets silence silly valgrind warnings, but we don't want
them in "production"
git-svn-id: svn://svn.handbrake.fr/HandBrake/trunk@4962 b64f7644-9d1e-0410-96f1-a4d463321fa5
|
|
|
|
|
|
|
|
| |
I missed a spot where an even wider left/right margin is
needed.
git-svn-id: svn://svn.handbrake.fr/HandBrake/trunk@4961 b64f7644-9d1e-0410-96f1-a4d463321fa5
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Use hb_buffer_t for reference buffers.
This is what eliminates extra buffer copies.
Simplified a lot of the code.
This resulted in some minor speed improvements and easier to read code.
Allow mcdeint+bob. Previously these could not be used together.
Thread the erode-dilate-erode-check steps in decomb3. More speed improvement.
Speed of default decomb went from 62fps to 76fps.
Speed of fast decomb went from 90fps to 95fps.
git-svn-id: svn://svn.handbrake.fr/HandBrake/trunk@4919 b64f7644-9d1e-0410-96f1-a4d463321fa5
|
|
|
|
|
|
|
|
|
|
| |
avpicture_deinterlace requires that both width and height of the input
be 8 pixel aligned. Video buffers already have padding to align horizontally
to 16 pixels, but they were not padded vertically. This adds vertical
padding.
git-svn-id: svn://svn.handbrake.fr/HandBrake/trunk@4776 b64f7644-9d1e-0410-96f1-a4d463321fa5
|
|
|
|
| |
git-svn-id: svn://svn.handbrake.fr/HandBrake/trunk@4737 b64f7644-9d1e-0410-96f1-a4d463321fa5
|
|
|
|
|
|
|
|
|
|
|
|
| |
when frames are duplicated, the frame was not copied properly and resulted
in divide by 0 later when the frame is processed.
This fix is a bit larger than the minimum size necessary to fix the problem.
I took the oportunity to clean up some buffer initialization and copy
operations.
git-svn-id: svn://svn.handbrake.fr/HandBrake/trunk@4563 b64f7644-9d1e-0410-96f1-a4d463321fa5
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
This patch enhances the filter objects. The 2 key improvements are:
1. A filter can change the image dimensions as frames pass through it.
2. A filter can output more than one frame.
In addition, I have:
Moved cropping & scalling into a filter object
Added 90 degree rotation to the rotate filter
Moved subtitle burn-in rendering to a filter object.
Moved VFR/CFR handling into a framerate shaping filter object.
Removed render.c since all it's responsibilities got moved to filters.
Improves VOBSUB and SSA subtitle handling. Allows subtitle animations.
SSA karaoke support.
My apologies in advance if anything breaks ;)
git-svn-id: svn://svn.handbrake.fr/HandBrake/trunk@4546 b64f7644-9d1e-0410-96f1-a4d463321fa5
|
|
|
|
|
|
|
| |
They are all ifdef'd out and not used anywhere. Enable and use as
needed when fifo issues crop up.
git-svn-id: svn://svn.handbrake.fr/HandBrake/trunk@4381 b64f7644-9d1e-0410-96f1-a4d463321fa5
|
|
|
|
|
|
|
|
|
| |
realloc doesn't really release memory under most circumstances, so it's
not suitable for reducing the size of an hb_buffer_t. So instead,
allocate a new smaller buffer, copy the data, and return the old larger
buffer to the buffer pool for reuse.
git-svn-id: svn://svn.handbrake.fr/HandBrake/trunk@4353 b64f7644-9d1e-0410-96f1-a4d463321fa5
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Specifically, this affects closed captions which are pushed into the
pipeline very late. CC's are dribbled in 2 bytes per video frame.
The entire CC that we put into a subtitle frame isn't available till
we see the signal for the end of the CC. This can be several seconds
after it began. So I have established a minimum buffer size that the
muxer accumulates before it outputs any data. This allows CCs to get
into the pipeline before the muxer outputs the interleave interval that
they belong in.
git-svn-id: svn://svn.handbrake.fr/HandBrake/trunk@4351 b64f7644-9d1e-0410-96f1-a4d463321fa5
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Adds support for MPEG-1 PS, HDDVD EVOB, and video codecs other
than mpeg1/2 in PS
Improves probing of unknown stream types by using Libav's probing
utilities
Use Libav to probe for dts profile in TS and PS files when profile is
unknown
Improves framerate detection (improved telecine detection)
Fixes preview generation for mpeg video that has only a single sequence
header
Patches Libav to handle VC-1 pulldown flags properly
Improve PS and TS stream log information
git-svn-id: svn://svn.handbrake.fr/HandBrake/trunk@4220 b64f7644-9d1e-0410-96f1-a4d463321fa5
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
TrueHD and DTS-HD now show up in the audio list along side their
AC-3 and DTS counterparts.
Note that currently the DTS-HD decoder we are using (ffmpeg) discards
the HD portion of the stream and onle decodes the DTS core portion. So
there is no advantage yet to using the DTS-HD stream. In the future
I would like to add DTS-HD passthru support and hopefully ffmpeg will
improve their DTS-HD decoder.
git-svn-id: svn://svn.handbrake.fr/HandBrake/trunk@3950 b64f7644-9d1e-0410-96f1-a4d463321fa5
|
|
|
|
|
|
|
| |
Makes it easier to read. Gets rid of some unnecessary variables.
git-svn-id: svn://svn.handbrake.fr/HandBrake/trunk@3886 b64f7644-9d1e-0410-96f1-a4d463321fa5
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Waiting for a fill threshhold in the fifos causes some non-determinism
in finding the first PTS value. Sometimes the fill level of one fifo
would not be reached until after another fifo is completely full, causing
an early exit in the loop that looks for the first PTS. When the initial PTS
is different between passes, the duration of the first frame is different.
This affects the PFR algorithm and can cause it to drop a different number
of frames.
The fill level was initially intended as a way to prevent thrashing between
threads to improve performance. But my testing indicates no degradation
when removing it.
git-svn-id: svn://svn.handbrake.fr/HandBrake/trunk@3819 b64f7644-9d1e-0410-96f1-a4d463321fa5
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
* New subtitle sync algorithm added to sync work-object ("simultaneous").
Classic algorithm preserved but disabled.
* Render work-object now supports queueing a /list/ of subtitles.
* FIFOs have been extended to support pushing/popping buffer-lists as single elements.
* Added SUBSYNC_VERBOSE_TIMING flag to debug timing issues related to subtitle display.
Observable behaviors changed in the new subtitle sync algorithm:
* Temporally overlapping subtitles are no longer trimmed to be non-overlapping.
* Subtitles less than two seconds long are no longer artificially extended. Sorry, Indochine fans.
* Subtitles that stop before they start will never be displayed. The old algorithm will display such subtitles if they begin in the future (relative to the current video frame being processed).
git-svn-id: svn://svn.handbrake.fr/HandBrake/trunk@3804 b64f7644-9d1e-0410-96f1-a4d463321fa5
|
|
|
|
| |
git-svn-id: svn://svn.handbrake.fr/HandBrake/trunk@3797 b64f7644-9d1e-0410-96f1-a4d463321fa5
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Unencrypted BD directory trees only. Doesn't support iso images.
Also, no PGS subtitle support yet.
Chapters and angles are supported.
Adds a new contrib libbluray.
Adds new option to hb_scan() for duration of short titles to filter.
This applies to BD and DVD multi-title scans only. Does not apply
to any single title scans.
Fixes memory leak during scan. hb_buffer_close() was not freeing
all buffers in a chain of buffers passed to it.
git-svn-id: svn://svn.handbrake.fr/HandBrake/trunk@3510 b64f7644-9d1e-0410-96f1-a4d463321fa5
|